What is the biggest blocker to trusting AI with customer-facing actions in your team?

This question keeps coming up in almost every product conversation we have.

For internal actions, the trust threshold is relatively low. Drafting a document, scoring a lead, updating a record, running an analysis. Teams are generally comfortable with AI handling these.

The threshold shifts the moment an action reaches a customer. Sending an email, publishing a post, running an ad, making an outbound call.

The blockers I hear most often:

- "What if it sends something off-brand or inaccurate?"

- "We have no audit trail for what the AI sent and why"

- "Legal needs to review anything customer-facing before it goes"

- "If it makes a mistake at scale, the damage is not proportional"

Which of these is the real blocker for your team, or is it something else entirely?